Machine Learning Predictive Models for Survival in Patients with Brain Stroke.

Background: This study aims to harness the predictive power of machine learning (ML) algorithms for accurately predicting mortality and survival outcomes in brain stroke (BS) patients. Methods: A total of 332 patients diagnosed with BS were enrolled in the study between April 21, 2006, and December 22, 2007, and then followed for 15 years (until 2023). Mortality outcomes were modeled using various statistical techniques, including the Cox model, decision trees, random survival forests (RSF), support vector machines (SVM), gradient boosting, and mboost. The best-performing model was selected based on diagnostic performance metrics: specificity, sensitivity, precision, accuracy, area under the receiver operating characteristic curve (AUC), positive likelihood ratio, negative likelihood ratio, and negative predictive value. Results: The results indicate that ML models in small sample sizes, particularly the SVM, outperformed the Cox model in predicting mortality and survival over 15 years, achieving an accuracy of 85% and an AUC of 0.765 (95% CI 0.637-0.83). Furthermore, the study identified important variables, including blood pressure history, waterpipe smoking, lack of physical activity, type of cerebrovascular accident, current smoking status, sex, and age, which provide valuable insights for clinicians in risk assessment. Conclusion: Our study showed that the SVM model outperforms the Cox model in predicting 15-year mortality and survival, particularly in small sample sizes. Moreover, the identification of key risk factors such as blood pressure history, waterpipe smoking, lack of physical activity, type of cerebrovascular accident, current smoking status, sex, and age highlights the need for their consideration in clinical assessments to enhance patient care.